Record successful public inference with custom audio references

Support user-provided speaker audio with a bundled reference encoder

Reorganize model card around training, samples, and usage

Present DuDE as full-duplex conversation TTS with named samples and verified held-out provenance

Record successful anonymous download and standalone GPU inference verification

Lead the model page with the two playable stereo results and expandable XML

Present the two featured audio recordings and XML; remove interactive demo

Include local Gradio app and standalone installation guide

Release standalone duplex TTS with XML interleaver, voice presets and examples

initial commit